New Workflow Separates AI-Generated API Docs from Human-Owned Commitments

A proposed documentation workflow splits API reference content into two distinct lanes: 'compile' lanes, where AI models restate facts already verified in a repository, and 'signature' lanes, reserved for human-authored commitments such as deprecation timelines, support terms, and security guarantees. The core problem it addresses is that mixing both types in a single AI-generated draft hides unverified promises inside text that reviewers may assume is factually checked. A decision table maps every documentation block type to its lane, allowed inputs, and a strict merge rule that fails if content is untagged or misclassified. A classifier tool runs before merge to enforce these rules, rejecting any signature-type statement that lacks a named human owner. The workflow treats the lane decision table as a formal policy document, meaning changes to it require the same review process as changes to an OpenAPI specification.